@uniqent/adapter-claude-code

Install a .uniqent agent brain into Claude Code.

Part of Uniqent — any brain, any agent: package an AI agent's whole brain into one open, signed .uniqent file and install it into whatever framework you run.

What it does

Translates a canonical .uniqent bundle into Claude Code's native layout:

• skills → .claude/skills/
• persona + policies + profile + memory facts → a <!-- uniqent:begin -->…<!-- uniqent:end --> managed section of AGENTS.md (your own content is preserved; an existing CLAUDE.md gets a managed @AGENTS.md import, since Claude Code ignores AGENTS.md when CLAUDE.md exists)
• MCP servers → merged .mcp.json (your credentials are resolved locally — secrets never travel in the bundle; a file holding resolved credentials is written with mode 0600 and must not be committed)
• channels / tasks / tools / runtime config / episodic memory / MCP tool allowlists that Claude Code can't hold are reported as lossiness, never silently dropped

Implements the @uniqent/adapter-sdk Adapter interface (detect / plan / apply / export).

Install

npm install @uniqent/adapter-claude-code

Most people don't need this directly. Use the CLI instead:

npx @uniqent/cli install my-brain.uniqent --target claude-code --root .

Usage (programmatic)

import { claudeCodeAdapter } from '@uniqent/adapter-claude-code';
import { readBundle } from '@uniqent/core';
import type { ResolvedCredentials } from '@uniqent/adapter-sdk';

// Load a signed .uniqent bundle
const bundle = await readBundle('./my-brain.uniqent');

// 1. Plan — inspect what will be written and what credentials are needed
const plan = await claudeCodeAdapter.plan(bundle, { root: '/path/to/project' });
console.log('Writes:', plan.writes);
console.log('Needs credentials:', plan.requiresCredentials);
console.log('Lossiness:', plan.lossiness);

// 2. Resolve credentials locally (never stored in the bundle)
const resolved: ResolvedCredentials = {
  'github-mcp_token': process.env.GITHUB_TOKEN ?? '',
};

// 3. Apply — write the native Claude Code layout into the target directory
const result = await claudeCodeAdapter.apply(bundle, plan, resolved, { root: '/path/to/project' });
console.log('Written:', result.written);
console.log('Notes:', result.notes);

Export (capture an existing Claude Code setup)

import { claudeCodeAdapter } from '@uniqent/adapter-claude-code';
import { packBundle } from '@uniqent/core';

// Reverse-translate a Claude Code project back into a portable .uniqent bundle.
// Credential values are scrubbed — only the requirement refs are captured.
const bundle = await claudeCodeAdapter.export({ root: '/path/to/project' });
await packBundle(bundle, './captured-brain.uniqent');

What gets written

| Bundle component | Claude Code target | Notes | | --------------------------- | ------------------------ | -------------------------------------------------------- | | Skills | .claude/skills/<name>/ | One directory per skill | | Persona + policies + memory | AGENTS.md | Memory folded in as flat text | | MCP servers | .mcp.json | Merged with any existing servers; creds injected locally | | Channels | — | Reported as lossiness (dropped) | | Tasks / automations | — | Reported as lossiness (dropped) | | Native tools | — | Reported as lossiness (dropped) |

apply is idempotent — running it twice produces the same result.

License

Apache-2.0. See the Uniqent monorepo.